[PATCH net v2] ipv4: fix conntrack defrag forwarding after ip_frag_reinit()

From: Jiayuan Chen

Date: Fri Nov 20 2026 - 04:47:36 EST


q.max_size and max_df_size are used by ip_frag_reasm() to set
IPCB(skb)->frag_max_size and the DF bit of the reassembled packet.

ip_frag_reinit() reuses the queue for a new datagram but does not
reset them, so the new datagram inherits them from the old one.

On a router with conntrack defrag, if the largest fragment of the old
datagram had DF set and was larger than the egress MTU, a new datagram
without DF gets DF set after reassembly and is rejected with
ICMP_FRAG_NEEDED, even though all its fragments fit.

Reset both fields.

Fixes: 5f2d04f1f9b5 ("ipv4: fix path MTU discovery with connection tracking")
Fixes: d6b915e29f4a ("ip_fragment: don't forward defragmented DF packet")
Reviewed-by: Eric Dumazet <edumazet@xxxxxxxxxx>
Signed-off-by: Jiayuan Chen <jiayuan.chen@xxxxxxxxx>
---
v1 -> v2: Target net
v1: https://lore.kernel.org/all/20261119061751.222641-1-jiayuan.chen@xxxxxxxxx/
---
net/ipv4/ip_fragment.c | 2 ++
1 file changed, 2 insertions(+)

diff --git a/net/ipv4/ip_fragment.c b/net/ipv4/ip_fragment.c
index ef7fd61284a3..0688be8142c4 100644
--- a/net/ipv4/ip_fragment.c
+++ b/net/ipv4/ip_fragment.c
@@ -250,6 +250,8 @@ static int ip_frag_reinit(struct ipq *qp)
qp->q.flags = 0;
qp->q.len = 0;
qp->q.meat = 0;
+ qp->q.max_size = 0;
+ qp->max_df_size = 0;
qp->iif = 0;
qp->ecn = 0;

--
2.43.0